An attribution model is the rule that decides which channel gets the credit for a conversion when someone has come through several sources before buying or filling in a form: a Google search, an ad, a newsletter, a direct visit. Change the rule and you change the channel that seems to have brought in the customer, and organic search is usually the one that loses most.
The guides on the first page of Google UK explain the models in the abstract or for paid campaigns, and some still list models Google removed in 2023. Here we look at them from the SEO side, with data from three Google Analytics 4 properties we manage: how much the model really shifts, and how much weight sits instead with everything Analytics can't see.
The attribution models, one by one
The models fall into two families. Rule-based models assign credit with a fixed formula someone has chosen. The data-driven model assigns it with an algorithm that compares the paths of people who convert with those of people who don't.
Take a typical B2B purchase path: the person finds a blog article on Google, a week later clicks a Google Ads ad, then comes back by typing the address and asks for a quote. Here's how each model splits that conversion, and what's left for SEO.
| Model | How it splits the credit | Organic search's share in the example | Available in GA4 today |
|---|---|---|---|
| Last click | 100% to the last channel clicked before the conversion | 0% (it all goes to Google Ads) | Yes, as "paid and organic last click" |
| First click | 100% to the first touchpoint | 100% | No, removed in 2023 |
| Linear | Equal shares to every touchpoint | 50% (the direct visit is excluded) | No, removed in 2023 |
| Time decay | More credit to touchpoints close to the conversion | Less than 50% | No, removed in 2023 |
| Position-based | 40% to the first, 40% to the last, 20% split among those in between | 50% with only two clicked touchpoints | No, removed in 2023 |
| Data-driven | Calculated by the algorithm on the property's real paths | Variable | Yes, it's the default |
| Google paid channels last click | 100% to the last Google ad, if there is one | 0% whenever a Google ad appears in the path | Yes |
The SEO column goes from zero to a hundred for exactly the same person. That's why the question "how many customers does SEO bring in?" has no single answer: it has one per model, and before comparing two reports you need to know which rule they were built with.
What Google removed in 2023
First click, linear, time decay and position-based no longer exist in Google's tools. In Google Ads they went in two stages: from June 2023 you could no longer choose them for new conversion actions, and from September 2023 the ones still using them were switched to the data-driven model. In Google Analytics 4 the documentation on attribution settings says they "are no longer available as of November 2023".
Today GA4 has three reporting models left: data-driven, paid and organic last click, and Google paid channels last click. On Google UK on 10 October 2026, the AI Overview for "attribution models" listed first touch, last touch, linear, time decay, position-based and data-driven without mentioning that Google has removed four of them. Among the organic results, Mediahawk describes the models without tying them to any platform, while Candid Digital's guide to Analytics attribution models, updated in March 2026, still lists seven, first click, linear, position-based and time decay included, as models to choose from. If you want a rule-based model today you have to build it outside Google, for example by exporting the data to BigQuery or using your CRM.
For SEO, losing first click hurts most. It was the only one-click model that rewarded informational content, the kind that introduces a business to someone who isn't yet searching for it by name.
How GA4 attributes an organic conversion today
Three rules in Google Analytics decide almost everything, and it's worth knowing them before you read an organic traffic report.
The first is about direct visits. According to Google's guide to attribution, "all attribution models exclude direct visits from receiving attribution credit", unless the path is made up entirely of direct visits. Someone who arrives from Google, leaves and comes back by typing the address is therefore assigned to organic search. It's a rule that helps SEO, and one many people forget when they blame "direct" for stealing conversions.
The second is the lookback window, meaning how far back in time a touchpoint can still get credit. For acquisition events (first_visit and first_open) it's 30 days, which you can cut to 7. For all other key events it's 90 days, which you can cut to 60 or 30. In B2B, with six-month sales cycles, an article read in January gets nothing for a contract signed in June.
The third is the reporting model. With the data-driven model, credit is split into fractions: it's normal to see 12.4 conversions attributed to organic search. With paid and organic last click, conversions are whole numbers and everything goes to the last channel clicked. You change the model in Admin, under Data display, in Attribution settings, and according to Google the change applies to both future and historical data: a report you exported last month may no longer match the same report opened today.
Why SEO looks less valuable with last click
Organic content does most of its work early in the path. Someone searches "how to cut compressor maintenance costs", reads an article, closes the page. Ten days later they search for the company's name, and above the organic result they find the ad from the brand campaign. They click that and fill in the form. With last click the conversion goes entirely to Google Ads, and the brand keyword looks like the most profitable one in the account.
It happens a lot in businesses that do both SEO and paid campaigns, and it's why the two channels are worth reading together, as we explain in our article on how SEO and Google Ads work together. There's also the opposite effect: when the brand search comes through organic, last click hands SEO conversions that may have started at a trade fair or through word of mouth. Brand searches should be read separately, because they measure awareness more than ranking work.
How much it really changes: three GA4 properties compared
We wanted to know how much the choice of model weighs on real sites, not on a textbook example. We pulled key events from GA4 from 1 October 2025 to 30 September 2026 for three properties: two client businesses in Italy, which we call site A and site B, and visilay.com. For each property we split the same key events in two ways: by the channel of the session in which they happened (the last touchpoint) and by the channel that first brought the user to the site (the first touchpoint, over the life of the cookie).
| Property | Key events in 12 months | Organic search, session channel | Organic search, first user channel | Difference |
|---|---|---|---|---|
| Site A (Italian business) | 544 | 165 (30.3%) | 164 (30.1%) | -0.2 points |
| Site B (Italian business, conversions mainly from Google Ads) | 2,297 | 455 (19.8%) | 441 (19.2%) | -0.6 points |
| visilay.com | 26 | 8 (30.8%) | 6 (23.1%) | -7.7 points |
On the two sites with meaningful volumes, moving from last to first touchpoint shifts organic search's share by less than one percentage point. On visilay.com the gap is wider, but with 26 key events in a year two conversions are enough to move the percentage by almost eight points: treat that one as an anecdote.
The reason the difference is so small, in our view, is that GA4 sees very short paths. To GA4 a user is a cookie: if they reject the banner, switch device or clear their browser data, the next visit is a new user, and their first touchpoint is the same as their last. The debate over which model to choose assumes long, well-tracked paths, which on most of the sites we work on the tool can't rebuild.
The problem comes before the model: the visits GA4 can't see
Every model works only on the visits Analytics records. To see how many are missing, we compared, over the same twelve months, web search clicks in Google Search Console, which Google counts on its side, with the google / organic sessions recorded by GA4.
| Property | Web search clicks (Search Console) | google / organic sessions (GA4) | Sessions / clicks ratio |
|---|---|---|---|
| Site A | 16,239 | 7,925 | 48.8% |
| visilay.com | 363 | 181 | 49.9% |
| Site B | 1,317 | 1,480 | 112.4% |
On two properties out of three, GA4 records about one organic visit for every two clicks Google counts. We call this number the organic coverage rate: the share of visits from Google that actually make it into Analytics, and the base every attribution model rests on. If it's 49%, half the paths are invisible before you even decide how to split the credit.
There are two main causes. The first is consent. Our three properties are tracked under Italian rules, where the Italian data protection authority's guidelines of 10 June 2021, in force since 9 January 2022, let third-party analytics cookies skip consent only under strict conditions; with GA4 set up normally, anyone who rejects the banner isn't tracked. The UK has moved the other way: since 5 February 2026, under the Data (Use and Access) Act 2025, cookies used only to collect statistics to improve a site can be set without consent, as long as visitors get clear information and a simple, free way to object (ICO guidance on the exceptions). The ICO adds that a third-party analytics provider has to act as your processor and can't use the data for other purposes such as advertising; it doesn't name any tool, so whether a given GA4 set-up qualifies is something to check with whoever handles your compliance. The second cause is one Google itself points to on its page on Search Console data discrepancies: tools like Analytics only count users with JavaScript enabled, the time zones differ (Search Console uses Pacific Time) and some data is deduplicated.
Site B goes the opposite way, with more sessions than clicks. We don't have a definite explanation: one click can generate several sessions if the person stays idle past the timeout and then carries on browsing, and a banner set-up that doesn't really block Analytics before consent would inflate the count. A ratio above 100% is a signal to check, not good news.
Google's consent mode can estimate the behaviour of people who decline, but only above thresholds a small B2B site rarely reaches. The documentation on behavioural modelling asks for at least 1,000 events a day with analytics_storage denied for at least 7 days, and at least 1,000 daily users with consent granted on 7 of the previous 28 days. If you're weighing up other tools, we've gathered the options and their limits in our guide to Google Analytics alternatives.
AI Overviews and AI assistants: the part of the path that leaves no trace
A growing part of SEO's work is used up inside the results page. The Pew Research Center watched the browsing of 900 US adults in March 2025: when an AI summary appeared, they clicked a traditional result in 8% of visits, against 15% when there was no summary. On the same effect, Ahrefs measured, across 300,000 informational keywords, a CTR about 34.5% lower for the top position when an AI Overview was present. Neither study covers the UK; what we've seen on real queries is in our article on the impact of AI Overviews on SEO.
For attribution the outcome is simple: if the person reads the answer and doesn't click, the first touchpoint doesn't exist for any model. When they search for the company by name ten days later, the conversion goes to brand search or to an ad. It's the same mechanism as zero-click searches, moved further upstream.
On the assistant side, Google has added a default "AI Assistant" channel to GA4, announced in the release notes of 13 May 2026, which groups visits from services such as ChatGPT, Gemini and Claude. On our properties, from June to September 2026, the ratio between AI Assistant sessions and organic sessions varies a lot from site to site.
| Property | Organic Search sessions | AI Assistant sessions | AI Assistant per 100 organic sessions |
|---|---|---|---|
| Site A | 2,038 | 43 | 2.1 |
| Site B | 666 | 79 | 11.9 |
| visilay.com | 73 | 10 | 13.7 |
The channel only counts visits that arrive with a referrer, though. Search Engine Journal points out two limits: assistants' mobile apps often pass no referrer at all, so those visits end up in Direct, while Google's AI Overviews and AI Mode stay inside organic search. To measure how often your brand appears in the answers, before any click, you need a different kind of tool, which we describe in our guide to AI visibility.
How to measure SEO's contribution without relying on one model
If the model shifts less than one point and the missing data is worth half, your time is better spent on what you can measure more reliably. These are the five things we set up, in this order.
- The organic coverage rate, every month. GA4 google / organic sessions divided by Search Console web clicks, over the same period. If it falls, the problem is tracking and no model will fix it. If it goes above 100%, check the cookie banner.
- The source saved on the contact, in the CRM. Hidden fields in the form record the first-visit source and the last one, and pass them to HubSpot, Pipedrive or Salesforce along with the lead. Tools such as Attributer or, on WordPress, Handl UTM Grabber do it without development work; you can also build it by hand with Google Tag Manager. The advantage is that the data follows the contact all the way to the contract, past GA4's 90-day window.
- The question "how did you hear about us?". An open field in the form, or a question from sales on the first call. The answers are imprecise, but they catch what no cookie sees: the ChatGPT answer, the article read on a phone and never clicked from the computer, the colleague who forwarded a link.
- Comparing two models, not choosing one. Putting the data-driven model next to paid and organic last click shows how much SEO contributes to paths someone else closes. If the gap is small, as in our data, it means GA4 sees short paths, not that SEO only works at the closing stage.
- Brand searches in Search Console. Impressions for queries containing the company name are the closest indicator of the effect of content that doesn't generate clicks. We read them alongside SEO KPIs and SEO ROI, never instead of them.
All of this ends up in a single dashboard, described in our article on the SEO dashboard, while which sources to trust when the numbers don't match is the subject of our guide to SEO analytics. If the problem is explaining it to whoever signs off the budget, we've also written about how to explain SEO to your boss.
The model really worth choosing
Marketing meetings spend a lot of time on which attribution model is fairest, and almost none on how many conversions the model gets to work with. On the sites we look after, the choice of model moves SEO's share by less than one point, while tracking loses half of it. The most useful model for a mid-sized business, in our view, is the one the sales team can check: the source written on the contact in the CRM, compared every month with what GA4 says. When the two versions disagree, the CRM is usually right. It's also how we read results in our SEO consultancy and Google Ads management projects, where the two channels have to be measured on the same sheet.
Frequently asked questions about attribution models
Three: data-driven, which is the default, paid and organic last click, and Google paid channels last click. First click, linear, time decay and position-based have not been available since November 2023.
It depends on the buying journey, but in GA4 the useful comparison is between the data-driven model and paid and organic last click. For long sales cycles it pays to save the first-visit and last-visit source in your CRM, because GA4's window stops at 90 days.
No, not in GA4's attribution reports. All models exclude direct visits from credit unless the path is made up entirely of direct visits: someone who arrives from Google and then comes back by typing the address is attributed to organic search.
Because they count different things. Search Console counts clicks on Google's side; GA4 counts sessions from browsers it can track, which excludes people who decline consent where it's required and anyone without JavaScript. On two of the three properties we analysed, GA4 records about one session for every two clicks.
No. Since May 2026 GA4 has a dedicated channel, AI Assistant, for visits that arrive from assistants such as ChatGPT, Gemini and Claude with a referrer. Visits from apps with no referrer end up in Direct, while Google's AI Overviews and AI Mode stay in organic search.
